Automated Extraction of Window-to-Wall Ratio from Street View Imagery for Urban-Scale Energy Modeling: A Case Study of Vienna

Abstract

Accurate estimation of building energy performance at the urban scale requires information from multiple domains; among these, detailed façade-related parameters, particularly the window-to-wall-ratio (WWR), play a critical role in determining solar gains, heat losses, and indoor lighting conditions. However, WWR data is rarely available at large scales due to the labor-intensive nature of manual data collection, often leading to simplified assumptions in Urban Building Energy Modeling (UBEM). This study proposes an automated workflow for extracting WWR values from Street View Imagery (SVI), integrating geospatial data processing, image extraction, rectification, and deep learning-based object detection. Using Open Government Data (OGD) and Google Street View (GSV) data, façade images are collected, stitched, and processed to generate fronto-parallel representations. The methodology is applied to the residential building stock owned by the City of Vienna (Gemeindebauten), serving as a case study. A real-time object detection system based on a convolutional neural network model is trained to detect façade elements, and WWR is computed through pixel-area ratios. The results are validated against manually annotated datasets using both detection metrics and regression-based comparison. Furthermore, extracted WWR values are analyzed according to construction periods and applied in UBEM simulations. This analysis shows that certain façade conditions observed in the SVIs, such as later renovations or changes at the ground-floor level, are not reflected in archetype-based WWR assumptions from the literature. To assess this impact quantitatively, UBEM simulations are compared using two WWR scenarios: one based on archetypal assumptions and one based on extracted WWR values. The difference between these scenarios is evaluated using percentage change, allowing the influence of WWR variation on the simulation results to be clearly expressed, particularly for cooling demand. The findings of this study suggest that data-driven façade representation is a key factor in enhancing the reliability of urban-scale energy modeling.